BiLinear-VGG16 细粒度图像分类

import numpy as np
from keras.models import Sequential, Model
from keras.layers import Input, Reshape, Lambda
from keras.layers import Dropout, Flatten, Dense
from keras.preprocessing.image import ImageDataGenerator
from keras.callbacks import ModelCheckpoint, EarlyStopping, ReduceLROnPlateau
from keras import optimizers
from keras.applications.vgg16 import VGG16
from keras import backend as K


def batch_dot(cnn_ab):
    return K.batch_dot(cnn_ab[0], cnn_ab[1], axes=[1, 1])


def sign_sqrt(x):
    return K.sign(x) * K.sqrt(K.abs(x) + 1e-10)


def l2_norm(x):
    return K.l2_normalize(x, axis=-1)


def bilinear_vgg16(size, class_num):
    input_tensor = Input(shape=(size[0], size[1], 3))
    model_vgg16 = VGG16(include_top=False, weights='imagenet',
                        input_tensor=input_tensor)
    cnn_out_a = model_vgg16.layers[-2].output
    cnn_out_shape = model_vgg16.layers[-2].output_shape
    cnn_out_a = Reshape([cnn_out_shape[1]*cnn_out_shape[2],
                         cnn_out_shape[-1]])(cnn_out_a)
    cnn_out_b = cnn_out_a
    cnn_out_dot = Lambda(batch_dot)([cnn_out_a, cnn_out_b])
    cnn_out_dot = Reshape([cnn_out_shape[-1]*cnn_out_shape[-1]])(cnn_out_dot)

    sign_sqrt_out = Lambda(sign_sqrt)(cnn_out_dot)
    l2_norm_out = Lambda(l2_norm)(sign_sqrt_out)
    output = Dense(class_num, activation='softmax')(l2_norm_out)
    model = Model(input_tensor, output)
    model.compile(loss='categorical_crossentropy', optimizer=optimizers.SGD(lr=1e-4, momentum=0.9, decay=1e-6),
                  metrics=['accuracy'])
    print(model.summary())
    return model_vgg16


if __name__ == '__main__':
    size=(320,320)
    class_num=5
    model_vgg16 = bilinear_vgg16(size,class_num)